Servers in Sweden have low utilization

[Hypertext1071 6]

Random question that I was always curious about, does anyone know why most of the servers in Sweden have such low utilization? Doesn't seem to happen for any other country.

[Tech Jedi Alex 1518]

Looking in Eddie, I can deduce a possible reason.

If the scoring rule is set to Speed, which is the default, only four servers actually get a non-zero score, putting only those four into consideration of the Connect to best server function. The client count reflects that.
I quick-tested a connection to Sweden on Android, and Copernicus was chosen to be the best server.. huh.

Also interesting: The first three are hosted by Altushost, Segin is Netrouting, rest seems to be Kustbandet. ISP might play a role here, too.

[Hypertext1071 6]

I think that would definitely explain it, I'm guessing most people use Eddie to connect to AirVPN servers, and pick whatever server has the highest score. I just read your other posts going into detail on how the score for speed was calculated. Since ping, load percentage and user percentage are all comparable, it seems like that "Penality" factor must be extremely high then for the servers located in Uppsala.

[Tech Jedi Alex 1518]

Doesn't take much for the score to be reduced to 0 stars, though. Certainly doesn't need the penalty value to be set  (it's set nowhere in SE).
I don't understand the "speed" scoring, anyway. Neither does it relate to theoretically achievable throughput (load), nor to a stable connection (clients). It's illogical for me why a server with 96% usage and 165 clients gets a similar score as a 28% usage server with a whopping 335 clients connected. The scoring by latency makes more sense in that those 23-37 ms servers get 5* and Menkab only 3-4. We really need better scoring rules…

[Staff 10397]

12 hours ago, Tech Jedi Alex said:

We really need better scoring rules…

Hello!

You may be right. According to your user feeling, what is the best selection of server using quick connection mode (i.e. you do not force a white list of any type) between Eddie Desktop, Eddie Android and AirVPN Suite (if you ran two or all of them)? And what is the software that achieves the best selection inside a single forced country (when the country offers multiple servers)?

Kind regards
 

[Tech Jedi Alex 1518]

1 hour ago, Staff said:

According to your user feeling, what is the best selection of server using quick connection mode (i.e. you do not force a white list of any type) between Eddie Desktop, Eddie Android and AirVPN Suite (if you ran two or all of them)?

I checked Eddie on Linux and Android in parallel, thrice, with some delay between them. Star ratings from Linux with Speed rule.

#1, Subra seems like the better one due to latency.
Linux: Subra 13ms 62% 119u (3-star)
Android: Taiyi 26 ms 65% 112u (0-star)

#2 Diphda seems slightly better due to latency and load.
Linux: Diphda 12ms 48% 128u (3-star)
Android: Taiyi 18ms 65% 112u (0-star)

#3 After a ~15min delay. Very similar results, but Linux chose again marginally better.
Linux: Garnet 13ms 56% 121u (3-star)
Android: Taiyi 13ms 58% 124u (0-star)

Android's fixation on Taiyi is interesting, but not a bad choice in all cases.
 

1 hour ago, Staff said:

And what is the software that achieves the best selection inside a single forced country (when the country offers multiple servers)?

#1 UK. Comparable. None is better than the other.
Linux: Chow 26ms 29% 104u
Android: Naos 24ms 35% 99u

#2 Sweden. All are bad choices, even if Linux chose marginally better. The Kustbandet servers are somehow ignored completely, there are no load and no users on them.
Linux: Norma 24ms 79% 167u
Android: Copernicus 25ms 86% 162u

#3 Japan. Iskandar is a slightly better choice.
Linux: Iskandar 234ms 52% 76u
Android: Albaldah 236ms 76% 85u

---

In regards to the scoring rules, an idea. A math formula incorporating clients, load and latency, times a modifier for the scoring rule. Something that maybe rates clients, load and latency = 0 best, so one can use the lowest result.
I'm not a math whiz, frankly, but maybe use a function where y grows exponentially. The parabola comes to mind, f(x) = x^2, where f(0) = 0. Drawing a bit in a graphical calculator, something like 0.0001 * x^2 * (modifier) * x looks promising. Calculate this for all three, sum the result, lowest is best server. Gives some flexibility in that it's easy to add further data points to the calculation, or a broader selection of modifiers. I think it would even enable users to set the modifiers themselves because all scores will be based on this parabola function.
Modifier can also be in front, (mod*0.0001) * x^2. Then you'd set it as integer (default 1), and rising modifiers cause the curve to steepen quicker.